How to Identify Great Candidates From Minimal Data

One executive recruiting firm we work with told us "Usually the best candidates don't have all the keywords in their profiles. In fact, they have no keywords. They just have the company name and their job title that they haven't updated in a year."

That observation captures a problem most recruiting tools ignore. Some of the candidates your team most wants to reach have thin profiles and minimal keyword footprints. The tools designed to find talent exclude them by default because they optimize for profile completeness rather than candidate quality.

Keyword-based search makes this worse the harder you try. And when public data is thin, most recruiting teams have no framework for what to evaluate instead. This article covers why strong candidates might have sparse profiles, why common workarounds fail, and what to look at when the profile itself gives you nothing to work with.

Why keyword-rich profiles often return average candidates instead of the best ones

Keyword-based recruiting tools parse a job description into tokens and match those tokens against candidate profiles. The more tokens that match, the higher the candidate ranks. This logic sounds reasonable until you watch it run on a real search.

One recruiting firm we spoke with described running a search for robotics engineers. The tool broke "GCP" into three separate keywords: G, C, and P. It pulled "engineer" as a standalone token. The result was a list where a hardware test engineer with eight months of internship experience ranked as a 100% match for a senior robotics role. The recruiter's response: "I could have done that myself."

The core issue is that keyword systems reward candidates who describe themselves well over candidates who have done the work but never bothered to write it up. A candidate who spends time optimizing their profile for recruiter search terms will rank higher than a candidate who spent that time building autonomous systems at a company whose work is classified or proprietary.

This problem gets worse with specificity. The same firm found that the longer and more detailed they made their search prompts, the lower the result quality dropped. Adding context about "hands-on experience training models that map perception to action" did not filter the pool down to qualified candidates. It created 70 keywords that diluted the match scoring until the results were random.

A randomized study of 37,000 applicants found that candidates advanced through AI-assisted pipelines (which go beyond keyword matching) passed final interviews at 54%, compared with 34% in traditional keyword-screening pipelines. The gap is large enough to reshape how recruiting teams think about sourcing, because keyword search biases results toward the wrong candidates by design. (For a breakdown of which AI sourcing tools actually go beyond keyword matching, we covered that separately.)

One recruiter at an executive search firm captured the frustration: "You slap a job description in and they just parse all the keywords and return candidates. I could have done that myself. But if there's some sort of intelligent layer where it can really understand what I'm trying to do versus just taking a job description and saying, here are five more different keywords for you, that's a game changer."

Why strong candidates often end up with sparse profiles

If keyword search over-rewards optimized profiles, it helps to understand why many strong candidates have thin ones in the first place. The reason is the same across every firm we spoke with. These candidates don't need to be found.

A senior engineer working on proprietary systems at a company known for strong technical hiring has no reason to maintain a detailed public profile. Their work cannot be open-sourced, and their employer's reputation already does the signaling. They receive enough inbound interest from their network that a keyword-optimized profile would add nothing to their career.

One recruiting firm's co-founder was explicit about this: "Some of the best engineers we hire, if you go to their GitHub, there's nothing. They've made no commits." He cautioned against over-indexing on public signals like repositories or publications. A fifth-listed author on a research paper may have contributed very little to the actual work. A candidate whose GitHub is full of tutorial projects and starred repositories may look active but have shipped nothing meaningful.

This matches the broader pattern from developer communities. On Hacker News, a hiring manager reported looking at "maybe two" GitHub profiles in hiring processes with "no correlation" to developer quality. Another commenter noted: "I'm sure it helps for some % of jobs. But as an employer I never look at it." Engineers working in data infrastructure, defense, healthcare, and financial services face the same constraint. Their best work is proprietary and will never appear on a public profile.

This creates a blind spot in every keyword-based tool. Candidates who show up in standard searches tend to be the ones with time, incentive, and permission to maintain public profiles. Many strong candidates are missing from those results because their skills are in high enough demand that they never needed to. We wrote a full guide on how to source passive candidates using signals rather than keyword presence.

Why lookalike search fails on one-in-a-thousand executive and specialist roles

A common workaround for keyword limitations is lookalike search: give the tool a strong candidate's profile and ask it to find similar people. This approach works for high-volume roles where there are many qualified candidates with roughly interchangeable skills, but it breaks on specialist and executive roles.

One firm's partners have a combined four decades of executive recruiting experience. Their assessment of lookalike search across every tool they have tried: "No one's nailed this. No one's done this well. And it seems like a really simple one."

The reason lookalike search fails on rare roles is that it assumes a representative reference candidate exists. For a senior robotics perception engineer who has shipped production autonomous systems, there may be 200 plausible candidates in a metro area, not 2,000. The reference profile the tool uses may itself be atypical. And the "similarity" the tool measures is keyword overlap, which brings back the same problem: it compares what candidates wrote rather than what they did.

Lookalike search also cannot reason about adjacent industries. A recruiter working specialist engineering roles described how they regularly pull candidates from medical robotics into aerospace, or from defense avionics into commercial robotics, because the underlying skills transfer. But if the company preset does not include medical device manufacturers, the tool will never find those candidates. As that recruiter noted: "It doesn't think about adjacent companies that could be a fit. For me, a lot of times I might go into the medical robotics industry and pull someone out of there just because it's very relevant."

Executive search firms have understood this for years. Research from Intellerati describes how executive search research goes well past profile matching. Firms build organizational charts to identify where candidates sit, verify claims through primary sources, and run confidential back-channel conversations to pre-reference candidates before outreach. A profile match cannot replicate this.

What signals predict candidate quality when the profile is sparse

When keyword presence is unreliable, recruiting teams need a different framework. The executive and specialist recruiters we spoke with converge on four signal types that matter more than profile keywords.

Employer trajectory is the primary signal

Where a candidate has worked, and what those companies do, tells you more than any self-reported skill list. A five-year tenure at a company known for building production autonomous systems is a stronger indicator of robotics capability than a profile that lists "robotics" as a keyword. The firms we spoke with confirmed this: the main evaluation criteria are "the person's education and employment history and what the company did, where the person has actually worked at."

You can enrich sparse profiles with company-level data to understand what a candidate's employers actually do, what technologies they work with, and how large their engineering teams are. When the profile gives you nothing, the employer context fills the gap.

Adjacent-industry thinking expands the pool

Instead of searching for candidates who match a preset list of target companies, strong recruiters reason about which adjacent industries produce transferable skills. Avionics engineers may be strong robotics candidates, while medical device firmware engineers may fit embedded systems roles. This kind of lateral reasoning requires company and industry data rather than keyword matching.

Public artifacts are supplementary, not primary

When a candidate does have public work that is relevant to the role, it adds signal. But it should be treated as, in one recruiter's words, "a cherry on top of a sundae." The absence of public artifacts should carry zero negative weight. Many of the strongest candidates cannot publish their work because it is proprietary, classified, or covered by non-disclosure agreements.

Recruiter calibration is the irreplaceable layer

Tools can find candidates, filter by employer trajectory, and score by proximity to target company profiles. But the final judgment call requires human context that no algorithm captures. One recruiter described reviewing a candidate who ranked as a 100% match by the tool but was clearly wrong for the role: "This is probably the perfect example of the judgment piece that a lot of these tools don't have. I understand why it would do that. But I wouldn't reach out to this person."

Recruiting teams can use people search APIs with 60+ filters to build candidate lists around employer, seniority, tenure, geography, and function rather than keyword matching, then layer human judgment on top for final calibration. For a walkthrough of how recruiting agencies wire this into their day-to-day workflow, see our guide on integrating a people search API into agency recruiting.

How a recruiting system should rank sparse-profile talent without defaulting to keyword presence

The firms we spoke with want more than a slightly better keyword search. They want a system that starts with what the company does, reasons about which adjacent companies produce transferable skills, and scores candidates on trajectory rather than token overlap.

Start with employer context rather than keyword tokens

The first filter should be: which companies do work that is relevant to this role? This requires structured company data, including industry classification, technology stack, team composition, and what the company actually builds, rather than a list of company names the recruiter already knows. The recruiter who described pulling candidates from medical robotics companies did it because they could reason about what those companies built. An automated system needs that same company-level intelligence to find candidates the recruiter would have found manually.

Build a disqualification layer alongside the matching layer

One of the most consistent complaints from the firms we interviewed was false positives where candidates ranked as 100% matches who were clearly wrong for the role. A hardware test engineer with eight months of internship experience should not rank the same as a senior robotics perception engineer. The system needs to actively disqualify candidates who lack the required seniority, tenure, or functional experience, even if their keyword overlap scores well.

Incorporate adjacent-company intelligence

The recruiter's suggestion was that an ideal system would recommend adjacent industries during a search: "Hey, I would actually recommend going adjacent to the defense area or the avionics space." This requires mapping company relationships, competitive landscapes, and technology overlap at a structured data level, so the system can suggest candidate pools the recruiter might not have considered.

Keep the human judgment gate

The firms we spoke with were realistic about what automation can and cannot do. Their goal was not to replace recruiter judgment but to scale it. As one partner put it: "If we can get to 50 to 60 people that I'm just like, these are great, that's a huge jump." The system should produce a shortlist small enough for a human to review every profile, with enough context attached (employer history, company description, tenure, education) that the recruiter can make a judgment call without needing to research each candidate independently.

For sparse-profile candidates, the gap between keyword and context-based search is even wider than the Talentprise study suggests, because keyword tools have nothing to match on at all. Context-based systems can still evaluate employer trajectory, tenure patterns, and company intelligence even when the profile itself is bare.

Conclusion

A sparse profile is not a disqualifier. It is a gap in data that your tools need to work around rather than give up on. Many strong candidates in executive and specialist roles have thin public profiles because they do not need to be found through them. Their employer reputation and network already do the work that a keyword-rich profile does for less experienced candidates.

Recruiting systems that rank on employer trajectory, adjacent-company intelligence, and structured candidate data can surface these candidates where keyword search cannot. The firms already doing this well have moved past keyword optimization entirely and are evaluating candidates on what they have done rather than what they have written.

If your team sources for executive, specialist, or technical roles where strong candidates often have thin profiles, Crustdata's recruiting data layer provides the employer context, people search filters, and company intelligence that keyword tools cannot.